Rewards in the era of agents: Operate, modernize, or transform?
Rewards is entering a period of redesign unlike anything the function has experienced in a generation. Budgets are moving into rewards technology, agentic AI is moving from demonstration to production in selected workflows, and the questions coming from the business are becoming more fundamental: What return are we getting from the function? Where is capacity being consumed? What should rewards professionals do when some of the work can be carried by machines?
For those who design how organizations value, price and reward work, the important question is not whether to adopt more technology. It is whether each rewards process is still producing the outcome the business and workforce need — and, if it is not, what kind of choice the situation demands. The choice is not between technology and no technology. It is whether to operate, modernize or transform, applied process by process.
Operate is a confirmation of the status quo. It means continuing to run a process because its outcome remains right and its methods remain fit for purpose. It is not neglect, and it is not the absence of a decision. It is what we conclude after checking that the business still needs what the process produces.
Modernize improves execution without changing the underlying outcome. It automates repeatable work, reduces administrative burden, improves speed and lowers cost. It is often the right answer when the rewards philosophy is sound, but delivery is the constraint.
Transform starts from a different premise. It asks whether the business model, workforce strategy or employee experience has moved beyond what the current rewards process was designed to provide. It rebuilds from first principles, using technology and AI as design inputs rather than as additions to an inherited model. Its ambition is not a leaner version of last year’s process. It is a different source of value.
These choices are not competing programs for the function as a whole. They are decisions to be made and remade across the portfolio. A function may operate one process, modernize another and transform a third in the same planning cycle. The discipline is to make the verdict explicit — and to revisit it when the outcome itself is in question, not merely when the process begins to feel dated.
Three forces make that discipline urgent
-
First
The capacity equation has changed. Research indicates that AI and automation could replace more than half of a rewards team’s workload, including routine employee inquiries and benefits administration. Yet 72% of US organizations describe their compensation function as primarily manual or only partially automated, and only 1% describe modernization as advanced. We therefore face two questions at once: what work can technology carry, and what should the people do with the capacity released? The second question is the more consequential one. -
Second
Agentic capability is arriving unevenly. Among organizations already using AI in compensation, job matching and leveling are the most common applications, at 55%, followed by market pricing recommendations at 38%. Salary increase recommendations — the capability closest to an actual pay decision — sit at just 11%. That ordering is not simply a measure of technical maturity. It reflects the harder questions of governance, decision rights, explainability and human judgment as agents move closer to decisions with real consequences for people’s pay. -
Third
The workforce is paying attention. Employee concern about job loss due to AI has grown sharply, while many employees believe leaders underestimate its emotional and psychological impact. For a function whose product is the employment deal, this is not a communications issue to be handled after deployment. Role clarity, reskilling, trust and adoption are part of the design of the rewards operating model itself.
Frame, Converge, Deliver
-
Frame
Establishes the direction before the build begins. It asks what the business demands of rewards, what technology can genuinely do today, how the future experience should work, what roles and skills will be required, and what the current state actually looks like in terms of architecture, data readiness and cost to serve. Demand comes before design. Evidence comes before ambition. Otherwise, we risk perfecting foundations for a destination we have not yet chosen. -
Converge
Reconciles the ideal with the possible. It turns the target experience into an operating model and a sequence of decisions about which work belongs with specialists, HR partners, platforms, automation or AI agents. Each process is tested against four questions: is its current outcome desirable, feasible, viable and legal? A narrower test then asks whether the work is verifiable, containable, frequent and sufficiently specified to be automated or assigned to an agent. Where those conditions do not hold, human judgment remains. An agent drafts; a person decides. -
Deliver
Proves the model on live work. It builds a first version with real cycles, real data and real users; establishes governance for bias, privacy and human oversight; equips and reskills the people whose work changes; and scales through measured waves. Deployment is a milestone. Adoption is the deliverable.
How we make progress
We should not begin with a technology catalogue. We should begin with an honest inventory of the work: the outcome each process produces, the capacity it consumes, the decisions it requires and the value it creates. Some processes will be right to continue operating. Some will be ready to modernize. A smaller number may warrant transformation. A caution from the field: skills-based automation is easier to demonstrate than to govern, and a compelling platform is not the same thing as a workable operating model.
The rewards function will be redesigned over the next several years. We are not bystanders to that redesign. We are among its architects. The organizations that navigate it well will not be those with the most tools, but those willing to make the choices visible, prove value on live work and carry their people through the change. The starting point is not a technology decision. It is the question that makes every subsequent decision possible: which processes are we operating, which are we modernizing, and which are we transforming?